Transformer-based models for speech synthesis are a class of neural network models that leverage the Transformer architecture to generate high-quality speech waveforms. The Transformer architecture, originally introduced for machine translation tasks, has proven to be highly effective in various natural language processing tasks, including speech synthesis. The Transformer architecture utilizes self-attention mechanisms to capture dependencies and relationships between different elements of the input sequence. It allows the model to attend to relevant parts of the input text during the decoding process, enabling it to generate coherent and contextually appropriate speech. Transformer-based models for speech synthesis have several advantages. They can handle long-range dependencies more effectively than traditional recurrent neural network (RNN)-based models, allowing for better modeling of the temporal structure in speech. Additionally, the parallelizable nature of the Transformer architecture enables faster training and inference compared to sequential RNN models. Examples of speech synthesis system architectures that use transformers are the [[FastSpeech architecture|FastSpeech]] and [[FastSpeech2 architecture|FastSpeech2]].